COMP in (parallel) grammar writing

Martin Forst

Abstract

Proceedings of LFG06; CSLI Publications On-line

This paper is a grammar writer's reaction to the 'COMP debate', which has been going on in the LFG community for more than a decade now. Taking as a starting point the work by Dalrymple and Lodrup (2000), Alsina et al. (2005) and Berman (2006), I first consider the question with respect to a German large-coverage LFG. I show that, in addition to the reasons put forth by Alsina et al. (2005) and Berman (2006), there are further reasons to reinterpret as OBLthetas (or OBJthetas) the arguments that Dalrymple and Lødrup (2000) analyze as COMPs in German, a language which they consider as 'mixed'. These have to do with COMPs subcategorized for by nouns and, to a lesser extent, with past participles of OBJ experiencer psych-verbs. I then present some data from Spanish, a 'non-mixed' language, and show that the distinction introduced in the reinterpretation of COMPs of German nouns has a correlate in Spanish. Furthermore, I point out how the reinterpretation of COMP can increase parallelism between grammars, an argument that needs to be taken with caution, of course, but in my opinion, does have its place in parallel grammar development. The final section explains why the linguistically more adequate analysis without COMP is also more attractive from the point of view of grammar engineering or, in other words, why the enhanced descriptive elegance of a grammar leads to improved efficiency in its processing. I report an 11% gain in processing time with a revised grammar compared to an equivalent grammar that makes use of COMP.